| dc.description |
Financial institutions play a vital role to enhance the socio-economic development of any country. Despite being a key player in socio-economic development, large group of people is still isolated in accessing formal financial services. Saving and Credit Cooperative Societies (SACCOS) are type of Microfinance Financial Institutions (MFIs) that came to solve the problem of isolating many people in attaining financial services. Loan default is a big challenge that causes weak performance of many SACCOS and eventually collapses.
In resolving the problem of loan default, this study analysed SACCOS credit rating in Tanzania by using machine learning approach. The secondary data of borrowers’ information from 2015 to 2019 were collected from KKKT Arusha Road SACCOS Ltd. The experiment was conducted in Anaconda environment with Python 3.8. Seventy percent (70%) of data were used for training and thirty percent (30%) for testing.
The predictive variable influencing SACCOS members’ credit ratings were analysed by using Random Forest and Logistic Regression algorithms. The foremost variables found by the Random Forest algorithm in influencing SACCOS members’ credit ratings were age, interest rate and membership years while the least variables were marital status and gender. The foremost factors found by Logistic Regression include age, loan period and interest rate while the least factors include membership years and marital status.
In evaluating the relationship between factors of SACCOS members and their associated credit rating the best results were obtained when the random forest algorithm was fitted with eleven features which include age, interest rate, membership years, loan amount, disbursement month, expired month, loan cycle, purpose, loan period, gender and marital status. The evaluation scores were 95%, 98%, 97%, 98% for Accuracy, Precision, Recall and F1-Score respectively. Also, for logistic regression, the best performance was obtained when the algorithm was fitted with three features which include loan period, interest rate and gender. The evaluation scores were 74%, 98%, 74%, 85% for Accuracy, Precision, Recall and F1-Score respectively. Overall machine learning provided the best results in analysing SACCOS credit ratings. |
|